Supplement of Biogeosciences, 16, 1211–1224, 2019https://doi.org/10.5194/bg-16-1211-2019-supplement© Author(s) 2019. This work is distributed underthe Creative Commons Attribution 4.0 License.

Supplement of

Dispersal distances and migration rates at the arctic treeline in Siberia –a genetic and simulation-based studyStefan Kruse et al.

Correspondence to: Stefan Kruse (stefan.kruse@awi.de)

The copyright of individual parts of the supplement might differ from the CC BY 4.0 License.

1

S1 Genetic analyses

Table S1. Summary table of the eight loci applied for the 11 subpopulations from the Taymyr Peninsula, sorted by decreasing population genetic differentiation value FST. Observed and expected heterozygosity are given by HO and HE, respectively.

No. Locus 1 Multiplex 2 TAG 3 Observed fragment length (bp) Number of alleles

1 bcLK253 1 Q3 211-247 16.99±0.39 2 Ld101 1 Q4 196-236 15.74±0.79 3 bcLK228 2 Q4 133-269 18.70±0.66 4 bcLK189 3 Q2 152-242 33.39±1.50 5 bcLK211 1 Q2 194-250 22.97±1.09 6 Ld42 3 Q4 187-201 7.86±0.35 7 bcLK056 2 Q1 154-256 31.79±1.05 8 bcLK263 2 Q2 198-280 39.77±0.96

1 Locus – marker names beginning ‘bcLK’ are developed by Isoda and Watanabe (2006) and those with ‘Ld’ by Wagner et al., (2012); 2 Multiplex – number indicates the three primer mixes applied in a simultaneous PCR; 3 TAG - TAG – tailing sequence at forward primer: Q1 = TGTAAAACGACGGCCAGT (Schuelke, 2000); Q2 = TAGGAGTGCAGCAAGCAT; Q3 = CACTGCTTAGAGCGATGC; Q4 = CTAGTTATTGCTCAGCGGT (Q2–Q4, after Culley et al. (2008))).

Figure S1. Fraction of missing alleles for each of three height classes – tree, sapling (Sapl), and seedling (Seed) (y-axis) and locus (x-axis) within each height class and the average value.

0.58 % 2.6 % 0.4 %

3.39 % 0.42 %

0.68 % 4.05 % 1.35 % 0.76 %

0.16 % 0.33 % 3.1 % 0.33 % 0.49 %Total

Seed

Sapl

Tree

K253

K211

d101

K056

K263

K228

K189

Ld42

Mean

Locus

Po

pu

latio

n

0

1

2

3

4Percent Missing

Percent missing data per locus and population of ghclas

2

Figure S2. Genotype accumulation curve showing convergence at 5-7 loci from which nearly all 601 tested individuals can be differentiated.

100%

0

100

200

300

400

500

600601

1 2 3 4 5 6 7

Number of loci sampled

Num

ber

of m

ulti

locus g

enoty

pes

Genotype accumulation curve for ghclas

3

S1.1 Allele diversity

S1.1.1 Introduction The number of alleles per loci was analysed separately in three height classes: ‘seedling’ <0.4 m, ‘sapling’ – taller than seedlings but <2 m, and ‘tree’ >2 m. For the analyses, we resampled the dataset to avoid errors introduced by sample size. This was achieved by constructing 100 datasets from 30 randomly selected individuals of each height class. To check whether the loci were under the null expectation of the Hardy-Weinberg equilibrium, 𝜒2-tests were performed on the observed allele frequencies (‘hw.test’-function in ‘pegas’-library version 0.9 (Paradis, 2010)).

To exclude errors introduced by clonal reproduction we used clone-censored datasets for the analyses. By using all eight loci we could distinguish between all genotyped individuals. We identified 601 separate individuals and 11 clones (Fig. 3a). The members of one genetically identical group were up to 30 m distant from each other; similar distances were found for black spruce stunted forms (Gamache et al., 2003; Laberge et al., 2000).

S1.1.2 Results The number of alleles per locus was nearly equal among all height classes with two exceptions at locus bcLK189 and bcLK263, at which the allele number was slightly smaller for seedlings. Individuals of all height classes showed significant heterozygote deficits with an observed mean of ~0.69 and an expected heterozygosity of ~0.86 (p<0.001, Table S2, Fig. S4). At two loci (bcLK253 and bcLK263) observed values were close to the expected ratio and thus did not differ significantly from the Hardy-Weinberg equilibrium (Table S2).

S1.1.3 Discussion The analysed tree stand is characterised by a high gene diversity (number of alleles and expected heterozygosity of ~86%) compared to other studies which used the same or parts of the same markers (Babushkina et al., 2016; Oreshkova et al., 2013; Pluess, 2011). Nevertheless, we observe a heterozygote deficit, which results in significant deviations from Hardy-Weinberg equilibrium, even though the analysed trees grew in a large area (one hectare). This was observed in the treeline area, spanning from dense forest to single-tree stands on the southern Taymyr Peninsula and which seems to be unaffected by the sampling area (Kruse et al., 2018). In general, this can be indicative of a higher degree of inbreeding among individuals and thus local recruitment outweighs immigration (Arenas et al., 2012; Hartl and Clark, 2007), although no straightforward pattern arises from the comparison of heterozygosity values (mean over all loci) among the three height classes (trees, saplings, seedlings). Nevertheless, in detail, the amount of alleles in seedlings is lower at two loci, for which also the observed heterozygosity is lower than for the other two height classes. This trend was expected for seedlings at all loci, because younger cohorts typically show depressed heterozygosity, caused by the higher probability of local reproduction (Addisalem et al., 2016; Moran and Clark, 2012). Subsequently, due to self-thinning, selection takes place, generally preferring fitter individuals – assuming heterozygotes are generally fitter (heterosis effect, for example Babushkina et al., 2016) one expects the older an individual is, the fitter it is compared to other competitors.

4

Table S2. Heterozygosity values for each locus by height class. The analyses are based on 100 resampled datasets, rarefied to 30 individuals.

Locus Trees Saplings Seedlings

HO [%] HE [%] HO [%] HE [%] HO [%] HE [%]

Ld101 55.9±8.5 77.4±4.8 53.8±8 79.7±3.8 60.8±8.2 78.8±3.6

bcLK056 62.5±8.9 91±1.4 55.2±8.2 90.1±1.6 64.2±7.4 91±1.1

bcLK189 79.1±6.8 88.8±2 70.2±7 89.3±1.6 72.8±6.5 88.3±1.5

bcLK211 68.6±7.2 88.5±2.5 63.7±6.7 89.4±1.9 66.4±6.7 89.2±1.8

bcLK228 70.8±7.8 87.9±1.6 68.5±7.5 88.4±1.3 65.4±7.5 89±1.4

bcLK253 80.1±8 83.8±2.5 80.3±6.1 83.8±2.6 81.2±6.4 83.3±2.7

bcLK263 90.1±4.8 93.8±0.8 89.6±4.5 93.8±0.7 86.3±5.3 92.6±0.9

Ld42 54.3±8.6 76±3.2 64.3±8.1 74.9±2.9 53±7.8 77.5±2.7

All 70.2±7.6 85.9±2.3 68.2±7 86.2±2.1 68.7±7 86.2±2

Figure S4. Left: Number of alleles, Right: Observed (HO) and expected (HE) heterozygosity. Based on a rarefied dataset of 30 individuals.

5

Figure S5. Number of offspring assigned to a single parent in three size classes. Filled circles: mean values.

Figure S6. For each genotyped individual sample the smallest number of different alleles to the other samples was binned into 0 to a maximum of 16 alleles.

6

S2 Model adaptation

S2.1 Program code adaptation

S2.1.1 Seed dispersal function improvements For each dispersed seed the wind direction is randomly drawn from vegetation period wind data of the year of its dispersal. The ballistic maximal flight distance 𝐸0 (Equation 1) is estimated by species-specific size parameters following the approach of Matlack (1987), where 𝑉ℎ is defined as the horizontal wind speed and is chosen corresponding to the wind direction in the model. The release height 𝐻𝑡 is estimated at 75% of the individual’s height. 𝑉𝑑 is the descent rate for seeds and is estimated for Larix gmelinii by a linear regression using species data from Matlack (1987). For species having wing-scales attached to the seeds, this rate can be calculated by 𝑉𝑑 =

0.0032 ∗ √𝑤 + 0.4807 and is 0.86 m s-1, with the wing loading 𝑤 (Matlack 1987) for L. gmelinii. The variable 𝑤 is calculated by dividing the average seed weight (in microdyne) of 3.5 mg (Heit and Eliason, 1940; Lukkarinen et al., 2009) by the propagule area of 0.2 cm2 (Fu et al., 1999).

𝐸0 = 𝑉ℎ𝐻𝑡

𝑉𝑑 (1)

This variable 𝐸0 controls the standard deviation of the Gaussian term in the dispersal function of the model which is named originally ‘width’ in Equation 5 in Kruse et al. (2016)), consisting of the two dispersal function terms

𝐷𝐿𝑔𝑎𝑢𝑠𝑠𝑖𝑎𝑛(𝑟𝑛) = √2 ∗ 𝐸02 ∗ −1 ∗ 𝑙𝑜𝑔(𝑟𝑛) and 𝐷𝐿𝑓𝑎𝑡−𝑡𝑎𝑖𝑙𝑒𝑑(𝑟𝑛) = 𝑟𝑛(−1∗(1+∝)), with 𝑟𝑛 – random number

uniformely distributed between 0 and 1, 𝑠𝑑𝑖𝑠𝑡 – distance parameter for fitting and ∝ - scaling parameter for the fat tail of the function:

𝐷𝐿𝑔𝑎𝑢𝑠𝑠𝑖𝑎𝑛,𝑓𝑎𝑡−𝑡𝑎𝑖𝑙(𝑟𝑛) = 𝑠𝑑𝑖𝑠𝑡 ∗ 0.5 ∗ ((0.5 ∗ 𝐷𝐿𝑔𝑎𝑢𝑠𝑠𝑖𝑎𝑛(𝑟𝑛)) + (2 ∗ 𝐷𝐿𝑓𝑎𝑡−𝑡𝑎𝑖𝑙𝑒𝑑(𝑟𝑛))) (2)

S2.1.2 Growth function The tree growth now depends only on July temperature, because climate-tree ring-width comparisons showed no significant influence of precipitation (Epp et al., 2018). With the species-specific linear regression coefficients

we estimate the simulated tree growth in a year by 𝑗𝑢𝑙𝑖𝑛𝑑𝑒𝑥 = (0.078

1+𝑒14,825−𝐽𝑢𝑙𝑦𝑡𝑒𝑚𝑝𝑒𝑟𝑎𝑡𝑢𝑟𝑒) + 0.108, which was

further processed to the scale factor 𝑤𝑒𝑎𝑡ℎ𝑒𝑟𝑓𝑎𝑐𝑡𝑜𝑟 =𝑗𝑢𝑙𝑖𝑛𝑑𝑒𝑥−𝑚𝑖𝑛𝑖𝑚𝑎𝑙𝑔𝑟𝑜𝑤𝑡ℎ

𝑚𝑎𝑥𝑖𝑚𝑎𝑙𝑔𝑟𝑜𝑤𝑡ℎ−𝑚𝑖𝑛𝑖𝑚𝑎𝑙𝑔𝑟𝑜𝑤𝑡ℎ.

S2.1.3 Active layer thickness influences mortality The influence of the active layer thaw depth on the diameter growth of larch trees is estimated based on the results of Nakai et al. (2008). It describes a linear relationship allowing 100% diameter growth at 100 cm thaw depth and only 10% when reaching 10 cm, which is the minimum value for L. gmelinii. The active layer thickness 𝐴𝐿𝑇 (Equation 3) is estimated in metres for each year with the Stefan Formula, following simplifications by Hinkel and Nicholas (1995). It is determined by soil property parameter 𝑓𝑒=0.050 (Global Land Cover Characterization, Zhang et al., 2005) and the cumulative sum of daily temperatures exceeding the freezing points 𝐷𝐷𝑇:

𝐴𝐿𝑇(𝑦𝑒𝑎𝑟) = 1.0 − 𝑓𝑒 ∗ √𝐷𝐷𝑇(𝑦𝑒𝑎𝑟) (3)

7

Table S4. Overview of model parameters and processes for L. gmelinii individuals that differ from the original version (Kruse et al. 2016).

Parameter Value and dimension

References

Growth

Quadratic term of the equation for diameter growth rate b -0.003 ln(cm) cm-² data-based estimate similar to Fyllas et al. (2010) Linear term of the growth function a 0.030 ln(cm) cm-1

Constant term of the growth function c -1.98 ln(cm)

Seed production, dispersal and establishment

Factor of seed productivity fS 8 literature-based estimate (Kruklis & Milyutin, 1977, cited in Abaimov, 2010)

Background germination rate fBackground Germination 0.01 estimated

Horizontal seed dispersal distance depended on actual wind, or for at wind speed of 10 km/h

𝐸0 variable, 60.1 m estimated after Matlack (1987)

Seed descent rate 𝑉𝑑 0.86 m s-1 estimated descent rate based on Matlack (1987) Mortality

Background mortality rate mBackground 0.0001 yr-1 data-based estimate

Current tree growth influence factor on tree mortality fGrowth Mortality 0.0 estimated

Weather influence factor on tree mortality fWeather Mortality 0.1 estimated

Density influence factor on tree mortality fDensity Mortality 2.0 estimated

Seed fertility 𝑎𝑔𝑒𝑚𝑎𝑥,𝑠𝑒𝑒𝑑𝑠,𝐿.𝑔𝑚𝑒𝑙𝑖𝑛𝑖𝑖 2 yrs Ban et al. (1998)

Mean temperature of the coldest month (January) at the border of the species‘ geographical range

𝑇𝑚𝑖𝑛,𝐿.𝑔𝑚𝑒𝑙𝑖𝑛𝑖𝑖 -45 °C Shugart et al. (1992)

Exponent scaling the height influence on tree mortality yexp 0.2 estimated

Weather processing

Exponent scaling the influence of surrounding density for a tree

𝑒𝑑𝑒𝑛𝑠𝑖𝑡𝑦,𝑏𝑎𝑠𝑎𝑙 𝑖𝑛𝑓𝑙𝑢𝑒𝑛𝑐𝑒 0.1 estimated

Exponent scaling the density value 𝑒𝑑𝑒𝑛𝑠𝑖𝑡𝑦,𝑡𝑟𝑒𝑒−𝑡𝑖𝑙𝑒 0.5 estimated

8

S2.2 Simulation results To fit the simulated seed effective dispersal distance to observations (Fig. 5) we explored potential settings (I) to decrease the amount of recruitment close to the mother trees, (II) to shift the effective dispersal peak by 2-3 m and (III) to increase the effective recruitment at medium distances (~30-40 m). Therefore, we tuned two kinds of processes: parameters that determine the seed dispersal (model code: bcdopwxyzDEFGHIJK, Table S5) and tree density and parameters that set the impact of the tree’s mortality (efghijklmnAC), or both (qrstuv) (details on individual adaptations in Table S5).

Of all 36 different simulations, some parameters decreased the amount of near mother effective seed dispersal (I) (cdo-zD-I) of which only (o-z) decreased the distance of up to 2 m based on the shifted dispersal function, while an increase in the distance parameter of the Gaussian-function peak improves the simulated function strongly (D-I) (Figure S8). Of these a shift towards farther distances and an increase in medium distances (II+III) was achieved with adaptations of the dispersal function only (oq-sx-zD-I), whereas the others only shifted the peak to ~5 m with a decrease for medium distances (pt-w). Nevertheless, the sum of squares of deviations from the observed pattern was improved in these candidates by a few sets (yzD-I) to within 66% to 82% of the reference run. The model performed best with parameter set “I” which is a combination of an adjusted dispersal function and increased seed production rate (Figure S7, Figure S8, Table S5). These sets increase the distance of dispersed seeds from the mother tree and the probability of a recruit growing at medium distances from its source was increased as well. Still the ratio of on-site recruitment was lower than observed (between 45.70 and 46.70 compared to observed 56.77%). This was improved by other simulations (qt-wJ) but their general performance (lower correlation coefficients, Table S5) was weaker than the reference simulation without parameter changes or adaptations of the model (a).

S2.3 Discussion of the simulation improvement We achieved a good fit when increasing the peak of the dispersal function in the model to longer distances. The models where the distance from the centre of the distribution was shifted to 4 m improved the simulated effective seed dispersal distances best (“I”). However, the ratio between on-site recruitment and introductions from the exterior is around 10% lower than observed. This was not improved by the best-performing parameter set, but could be improved when changing the density competition, especially for small life stages (“t”). Combinations of both were tested but results strongly deviated from observed situations (“C”). This smaller ratio points to an unrealistically high long-distance dispersed seed fraction. Here, we focus on the effective seed dispersal distance at short distances. Nevertheless, long-distance dispersal should be improved too, especially if one aims to conduct simulation studies over larger extents. It could be improved by decreasing the fat tail probability of the exponential part in the dispersal function, or by manipulating the implementation of the wind speed influence to a nonlinear process, decreasing the distances for strong winds. We analysed only one area at the treeline, which improved our understanding of the processes incorporated in the simulation model, but this may overemphasise the effective seed dispersal of one subpopulation. Therefore further validation by more plot-based analyses is needed for the general function parameters.

9

Figure S7. Deviations of simulated versus observed effective seed dispersal. SS – sum of squares, R2 – square of correlation between the mean simulated series and the observed value, p – significance of correlation coefficient. Letters (a-zA-K) refer to a special simulation run s. Table S5 for details.

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

a

R2=0.7258 p=0.0000

N=10 SS=0.0096

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

b

R2=0.7187 p=0.0000

N=10 SS=0.0099

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

c

R2=0.7445 p=0.0000

N=10 SS=0.0092

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

d

R2=0.6279 p=0.0000

N=10 SS=0.0132

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

e

R2=0.6477 p=0.0000

N=10 SS=0.0125

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

f

R2=0.3310 p=0.0000

N=2 SS=0.0392

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

g

R2=0.6307 p=0.0000

N=10 SS=0.0131

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

h

R2=0.6884 p=0.0000

N=10 SS=0.0110

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

i

R2=0.7041 p=0.0000

N=10 SS=0.0104

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

j

R2=0.7331 p=0.0000

N=10 SS=0.0094

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

k

R2=0.7244 p=0.0000

N=10 SS=0.0098

0.5 2.0 10.0 50.0-0

.04

0.0

2

mids

density

l

R2=0.7170 p=0.0000

N=10 SS=0.0099

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

m

R2=0.7198 p=0.0000

N=10 SS=0.0098

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

n

R2=0.7210 p=0.0000

N=10 SS=0.0098

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

o

R2=0.7039 p=0.0000

N=10 SS=0.0136

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

p

R2=0.4955 p=0.0000

N=10 SS=0.0394

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

q

R2=0.7368 p=0.0000

N=10 SS=0.0115

0.5 2.0 10.0 50.0-0

.04

0.0

2

mids

density

r

R2=0.6624 p=0.0000

N=9 SS=0.0150

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

s

R2=0.7327 p=0.0000

N=10 SS=0.0110

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

t

R2=0.2971 p=0.0000

N=10 SS=0.0740

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

u

R2=0.3355 p=0.0000

N=10 SS=0.0753

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

v

R2=0.3289 p=0.0000

N=10 SS=0.0754

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

w

R2=0.3038 p=0.0000

N=10 SS=0.0838

0.5 2.0 10.0 50.0

-0.0

40.0

2mids

density

x

R2=0.7601 p=0.0000

N=10 SS=0.0113

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

y

R2=0.8321 p=0.0000

N=10 SS=0.0073

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

z

R2=0.8226 p=0.0000

N=10 SS=0.0079

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

A

R2=0.6745 p=0.0000

N=10 SS=0.0115

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

C

R2=0.2825 p=0.0000

N=5 SS=0.0293

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

D

R2=0.8094 p=0.0000

N=10 SS=0.0081

0.5 2.0 10.0 50.0

-0.0

40.0

2mids

density

E

R2=0.8141 p=0.0000

N=10 SS=0.0066

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

F

R2=0.8512 p=0.0000

N=10 SS=0.0068

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

G

R2=0.8186 p=0.0000

N=10 SS=0.0084

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

H

R2=0.8287 p=0.0000

N=10 SS=0.0079

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

I

R2=0.8588 p=0.0000

N=10 SS=0.0063

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

J

R2=0.0688 p=0.0084

N=9 SS=0.1623

0.5 2.0 10.0 50.0

-0.0

40.0

2

mids

density

K

R2=0.5282 p=0.0000

N=10 SS=0.0176

Fre

qu

en

ce

Distance to mother [m]

10

Figure S8. Deviations of effective seed dispersal distances of all runs from the reference simulation "a". Grey areas are the standard deviation of all runs. Red and blue dots indicate values outside (above and below respectively) the standard deviation of the base run. Letters (a-zA-K) refer to a special simulation run s. Table S5 for details.

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

a

R2=1.0000 p=0.0000

N=10 SS=0.0000

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

b

R2=0.9316 p=0.0000

N=10 SS=0.0019

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

c

R2=0.8601 p=0.0000

N=10 SS=0.0038

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

d

R2=0.6657 p=0.0000

N=10 SS=0.0089

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

e

R2=0.9534 p=0.0000

N=10 SS=0.0013

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

f

R2=0.6786 p=0.0000

N=2 SS=0.0183

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

g

R2=0.9446 p=0.0000

N=10 SS=0.0015

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

h

R2=0.9480 p=0.0000

N=10 SS=0.0014

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

i

R2=0.9543 p=0.0000

N=10 SS=0.0012

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

j

R2=0.9652 p=0.0000

N=10 SS=0.0010

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

k

R2=0.9676 p=0.0000

N=10 SS=0.0009

0.5 2.0 10.0 50.0-0

.04

0.0

2

h1x$mids

densm

eanvgl

l

R2=0.9592 p=0.0000

N=10 SS=0.0011

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

m

R2=0.9794 p=0.0000

N=10 SS=0.0006

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

n

R2=0.9735 p=0.0000

N=10 SS=0.0007

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

o

R2=0.8561 p=0.0000

N=10 SS=0.0077

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

p

R2=0.6938 p=0.0000

N=10 SS=0.0286

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

q

R2=0.8857 p=0.0000

N=10 SS=0.0061

0.5 2.0 10.0 50.0-0

.04

0.0

2

h1x$mids

densm

eanvgl

r

R2=0.8704 p=0.0000

N=9 SS=0.0066

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

s

R2=0.8906 p=0.0000

N=10 SS=0.0051

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

t

R2=0.5596 p=0.0000

N=10 SS=0.0526

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

u

R2=0.5578 p=0.0000

N=10 SS=0.0578

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

v

R2=0.5636 p=0.0000

N=10 SS=0.0568

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

w

R2=0.5223 p=0.0000

N=10 SS=0.0651

0.5 2.0 10.0 50.0

-0.0

40.0

2h1x$mids

densm

eanvgl

x

R2=0.8179 p=0.0000

N=10 SS=0.0096

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

y

R2=0.7949 p=0.0000

N=10 SS=0.0094

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

z

R2=0.7241 p=0.0000

N=10 SS=0.0126

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

A

R2=0.9516 p=0.0000

N=10 SS=0.0013

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

C

R2=0.6108 p=0.0000

N=5 SS=0.0118

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

D

R2=0.7872 p=0.0000

N=10 SS=0.0094

0.5 2.0 10.0 50.0

-0.0

40.0

2h1x$mids

densm

eanvgl

E

R2=0.8730 p=0.0000

N=10 SS=0.0039

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

F

R2=0.7791 p=0.0000

N=10 SS=0.0106

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

G

R2=0.7833 p=0.0000

N=10 SS=0.0107

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

H

R2=0.7711 p=0.0000

N=10 SS=0.0111

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

I

R2=0.7948 p=0.0000

N=10 SS=0.0097

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

J

R2=0.3084 p=0.0000

N=9 SS=0.1199

0.5 2.0 10.0 50.0

-0.0

40.0

2

h1x$mids

densm

eanvgl

K

R2=0.5519 p=0.0000

N=10 SS=0.0143

Fre

qu

en

ce

Distance to mother [m]

11

Table S5. Results of effective seed dispersal in adapted simulations. Changes identifier: 1-dispersal function manipulation, 2-density calculation manipulation, 3-combinations of 1 and 2.

Dispersal function On-site recruitment ratio Model parameters

ID Changes SS r r2 Ratio SD N>10 in center

p diff from obs

Adaptation and expected outcome Parameter1

a - 0.0096 0.8519 0.7258 46.8% 1.3% 10 0.0000 - reference run - Kruse et al. (2016) b 1 0.0099 0.8478 0.7187 45.4% 1.4% 10 0.0000 longer dispersal distances Sdist=1 c 1 0.0092 0.8629 0.7445 41.0% 1.2% 10 0.0000 longer dispersal distances Sdist=5 d 1 0.0132 0.7924 0.6279 38.1% 1.0% 10 0.0000 longer dispersal distances Sdist=10 e 2 0.0125 0.8048 0.6477 46.3% 1.7% 10 0.0000 larger distance to mother trees 𝑓𝐷𝑒𝑛𝑠𝑖𝑡𝑦𝑀𝑜𝑟𝑡𝑎𝑙𝑖𝑡𝑦=3

f 2 0.0392 0.5754 0.3310 45.6% 6.7% 3 0.1023 smaller distance to mother trees 𝑓𝐻𝐴𝐼=5 g 2 0.0131 0.7942 0.6307 44.8% 2.3% 10 0.0000 larger distance to mother trees 𝑓𝐻𝐴𝐼=15 h 2 0.0110 0.8297 0.6884 45.9% 2.6% 10 0.0000 less exclusion close to mother tree 𝑒𝑑𝑒𝑛𝑠𝑖𝑡𝑦,𝑏𝑎𝑠𝑎𝑙 𝑖𝑛𝑓𝑙𝑢𝑒𝑛𝑐𝑒=0.05

i 2 0.0104 0.8391 0.7041 48.4% 2.1% 10 0.0000 higher exclusion close to mother tree 𝑒𝑑𝑒𝑛𝑠𝑖𝑡𝑦,𝑏𝑎𝑠𝑎𝑙 𝑖𝑛𝑓𝑙𝑢𝑒𝑛𝑐𝑒=0.15

j 2 0.0094 0.8562 0.7331 46.2% 1.7% 10 0.0000 higher exclusion close to mother tree densitysmallweighing=1 k 2 0.0098 0.8511 0.7244 47.0% 1.2% 10 0.0000 less exclusion close to mother tree densitytreetile=0 l 2 0.0099 0.8467 0.7170 51.5% 2.2% 10 0.0000 higher exclusion close to mother tree densitytreetile=1

m 2 0.0098 0.8484 0.7198 46.9% 0.7% 10 0.0000 higher exclusion close to mother tree densitytiletree=1 n 2 0.0098 0.8491 0.7210 47.7% 1.3% 10 0.0000 higher exclusion close to mother tree densitymaxreduction=1 o 1 0.0136 0.8390 0.7039 47.9% 2.2% 10 0.0000 more distant from centre and more intense peak Sdist=1 + 𝑟𝐺𝑎𝑢𝑠𝑠𝐸𝑥𝑝𝐷𝑖𝑠𝑝=1.0 + 𝑑𝐺𝑎𝑢𝑠𝑠𝐶𝑒𝑛𝑡𝑟𝑒=2.0

p 1 0.0394 0.7039 0.4955 49.9% 3.2% 10 0.0001 shorter dispersal distances o + 𝑑𝐺𝑎𝑢𝑠𝑠𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒=𝐷 ∗ 0.5 q 3 0.0115 0.8584 0.7368 52.0% 2.0% 10 0.0000 higher exclusion close to mother tree o +

𝑒𝑑𝑒𝑛𝑠𝑖𝑡𝑦,𝑡𝑟𝑒𝑒−𝑡𝑖𝑙𝑒=1

r 3 0.0150 0.8139 0.6624 49.3% 2.8% 9 0.0000 less exclusion close to mother tree o + 𝑒𝑑𝑒𝑛𝑠𝑖𝑡𝑦,𝑏𝑎𝑠𝑎𝑙 𝑖𝑛𝑓𝑙𝑢𝑒𝑛𝑐𝑒=0.15

s 3 0.0110 0.8560 0.7327 44.7% 1.2% 10 0.0000 higher exclusion close to mother tree o + 𝑒𝑑𝑒𝑛𝑠𝑖𝑡𝑦,𝑡𝑖𝑙𝑒−𝑡𝑟𝑒𝑒=3

t 3 0.0699 0.5795 0.3358 52.4% 4.4% 10 0.0113 increased tree density o + treedensity^0.9 u 3 0.0706 0.5957 0.3548 50.9% 2.1% 10 0.0000 increased tree density o + treedensity^0.95 v 3 0.0788 0.5627 0.3166 51.0% 1.3% 5 0.0006 weakened tree density o + treedensity^1.1 w 1 0.0838 0.5512 0.3038 52.0% 2.0% 10 0.0000 shortened dispersal distance o + 𝑑𝐺𝑎𝑢𝑠𝑠𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒=𝐷0.5 x 1 0.0113 0.8718 0.7601 47.7% 1.4% 10 0.0000 more distant from centre and more intense peak Sdist=1 + 𝑟𝐺𝑎𝑢𝑠𝑠𝐸𝑥𝑝𝐷𝑖𝑠𝑝=1.0 + 𝑑𝐺𝑎𝑢𝑠𝑠𝐶𝑒𝑛𝑡𝑟𝑒 =3.0

y 1 0.0073 0.9122 0.8321 47.4% 1.3% 10 0.0000 more distant from centre and more intense peak Sdist=1 + 𝑟𝐺𝑎𝑢𝑠𝑠𝐸𝑥𝑝𝐷𝑖𝑠𝑝=1.0 + 𝑑𝐺𝑎𝑢𝑠𝑠𝐶𝑒𝑛𝑡𝑟𝑒 =4.0

z 1 0.0079 0.9070 0.8226 46.7% 2.2% 10 0.0000 more distant from centre and more intense peak Sdist=1 + 𝑟𝐺𝑎𝑢𝑠𝑠𝐸𝑥𝑝𝐷𝑖𝑠𝑝=1.0 + 𝑑𝐺𝑎𝑢𝑠𝑠𝐶𝑒𝑛𝑡𝑟𝑒 =5.0

A 2 0.0115 0.8213 0.6745 48.6% 2.9% 10 0.0000 higher exclusion close to mother tree linear density 0-200 cm 1-0 extra mortality C 2 0.0293 0.5315 0.2825 50.2% 7.5% 7 0.0613 higher exclusion close to mother tree negative quadratic density 0-200 cm 5-0 extra

mortality D 1 0.0081 0.8997 0.8094 45.9% 0.9% 10 0.0000 increased seed production higher on-site reproduction y + f𝑠=16 (twice standard) E 1 0.0066 0.9023 0.8141 45.7% 1.8% 10 0.0000 more distant shifted dispersal peak o + 𝑑𝐺𝑎𝑢𝑠𝑠𝐷𝑖𝑠𝑡𝑎𝑛𝑐𝑒=𝐷 ∗ 1.5 F 1 0.0068 0.9226 0.8512 46.2% 1.2% 10 0.0000 increased seed production higher on-site reproduction y + f𝑠=12 G 1 0.0084 0.9048 0.8186 46.6% 1.6% 10 0.0000 increased seed production higher on-site reproduction y + f𝑠=9 H 1 0.0079 0.9103 0.8287 45.7% 1.2% 10 0.0000 increased seed production higher on-site reproduction y + f𝑠=10 I 1 0.0063 0.9267 0.8588 46.4% 1.7% 10 0.0000 increased seed production higher on-site reproduction y + f𝑠=11 J 1 0.1623 0.2624 0.0688 54.9% 4.0% 9 0.2023 higher on-site reproduction a + no exponential dispersal K 1 0.0176 0.7268 0.5282 44.8% 1.3% 10 0.0000 higher on-site reproduction I + no exponential dispersal

1 – abbreviations following Kruse et al. (2016), Epp et al. (2018), and, Kruse et al. (2018)

12

References Abaimov, A. P.: Permafrost Ecosystems, edited by A. Osawa, O. A. Zyryanova, Y. Matsuura, T. Kajimoto, and R. W. Wein, Springer Netherlands, Dordrecht., 2010.

Addisalem, A. B., Duminil, J., Wouters, D., Bongers, F. and Smulders, M. J. M.: Fine-scale spatial genetic structure in the frankincense tree Boswellia papyrifera (Del.) Hochst. and implications for conservation, Tree Genet. Genomes, 12(5), 86, doi:10.1007/s11295-016-1039-2, 2016.

Arenas, M., Ray, N., Currat, M. and Excoffier, L.: Consequences of range contractions and range shifts on molecular diversity, Mol. Biol. Evol., 29(1), 207–218, doi:10.1093/molbev/msr187, 2012.

Babushkina, E. A., Vaganov, E. A., Grachev, A. M., Oreshkova, N. V., Belokopytova, L. V., Kostyakova, T. V. and Krutovsky, K. V.: The effect of individual genetic heterozygosity on general homeostasis, heterosis and resilience in Siberian larch (Larix sibirica Ledeb.) using dendrochronology and microsatellite loci genotyping, Dendrochronologia, 38, 26–37, 2016.

Ban, Y., Xu, H., Bergeron, Y. and Kneeshaw, D. D.: Gap regeneration of shade-intolerant Larix gmelini in old-growth boreal forests of northeastern China, J. Veg. Sci., 9(4), 529–536, doi:10.2307/3237268, 1998.

Culley, T. M., Weller, S. G., Sakai, A. K. and Putnam, K. A.: Characterization of microsatellite loci in the Hawaiian endemic shrub Schiedea adamantis (Caryophyllaceae) and amplification in related species and genera, Mol. Ecol. Resour., 8(5), 1081–1084, doi:10.1111/j.1755-0998.2008.02161.x, 2008.

Epp, L. S., Kruse, S., Kath, N. J., Stoof-Leichsenring, K. R., Tiedemann, R., Pestryakova, L. A. and Herzschuh, U.: Temporal and spatial patterns of mitochondrial haplotype and species distributions in Siberian larches inferred from ancient environmental DNA and modeling, Sci. Rep., 8(1), 17436, doi:10.1038/s41598-018-35550-w, 2018.

Fu, L., Li, N. and Mill, R. R.: Sections on Cephalotaxaceae, Ginkgoaceae and Pinaceae, in Flora of China, edited by W. Zheng-yi and P. H. Raven, pp. 11–52, Science Press; St. Louis: Missouri Botanical Garden, Beijing., 1999.

Fyllas, N. M., Politi, P. I., Galanidis, A., Dimitrakopoulos, P. G. and Arianoutsou, M.: Simulating regeneration and vegetation dynamics in Mediterranean coniferous forests, Ecol. Model., 221(11), 1494–1504, doi:10.1016/j.ecolmodel.2010.03.003, 2010.

Gamache, I., Jaramillo-Correa, J. P., Payette, S. and Bousquet, J.: Diverging patterns of mitochondrial and nuclear DNA diversity in subarctic black spruce: imprint of a founder effect associated with postglacial colonization., Mol. Ecol., 12(4), 891–901 [online] Available from: http://www.ncbi.nlm.nih.gov/pubmed/12753210, 2003.

Hartl, D. L. and Clark, A. G.: Principles Of Population Genetics, 4th ed., Sinauer Associates, Inc. Publishers, Sunderland, Massachusetts, USA., 2007.

Heit, E. C. and Eliason, E. J.: Coniferous tree seed testing and factors affecting germination and seed quality, Tech. Bull. New York State Agric. Exp. Stn., 255, 1–45, 1940.

Hinkel, K. M. and Nicholas, J. R. J.: Active layer thaw rate at a boreal forest site in central Alaska, USA, Arct. Alp.

13

Res., 1, 72–80, 1995.

Isoda, K. and Watanabe, A.: Isolation and characterization of microsatellite loci from Larix kaempferi, Mol. Ecol. Notes, 6(3), 664–666, doi:10.1111/j.1471-8286.2006.01291.x, 2006.

Kruse, S., Wieczorek, M., Jeltsch, F. and Herzschuh, U.: Treeline dynamics in Siberia under changing climates as inferred from an individual-based model for Larix, Ecol. Model., 338, 101–121, doi:10.1016/j.ecolmodel.2016.08.003, 2016.

Kruse, S., Epp, L. S., Wieczorek, M., Pestryakova, L. A., Stoof-Leichsenring, K. R. and Herzschuh, U.: High gene flow and complex treeline dynamics of Larix Mill. stands on the Taymyr Peninsula (north-central Siberia) revealed by nuclear microsatellites, Tree Genet. Genomes, 14(2), 19, doi:10.1007/s11295-018-1235-3, 2018.

Kruse, S., Gerdes, A., Kath, N. J. and Herzschuh, U.: Implementing spatially explicit wind-driven seed and pollen dispersal in the individual-based larch simulation model: LAVESI-WIND 1.0, Geosci. Model Dev., 11(11), 4451–4467, doi:10.5194/gmd-11-4451-2018, 2018.

Laberge, M.-J., Payette, S. and Bousquet, J.: Life span and biomass allocation of stunted black spruce clones in the subarctic environement, J. Ecol., 88, 584–593, 2000.

Lukkarinen, A. J., Ruotsalainen, S., Nikkanen, T. and Peltola, H.: The Growth Rhythm and Height Growth of Seedlings of Siberian (Larix sibirica Ledeb.) and Dahurian (Larix gmelinii Rupr.) Larch Provenances in Greenhouse Conditions, Silva Fenn., 43(1), 5–20, 2009.

Matlack, G. R.: Diaspore size, shape, and fall behavior in wind-dispersed plant species, Am. J. Bot., 74(8), 1150–1160, doi:10.2307/2444151, 1987.

Moran, E. V. and Clark, J. S.: Between-site differences in the scale of dispersal and gene flow in red oak, PLoS One, 7(5), e36492, doi:10.1371/journal.pone.0036492, 2012.

Nakai, Y., Matsuura, Y., Kajimoto, T., Abaimov, a. P., Yamamoto, S. and Zyryanova, O. a.: Eddy covariance CO2 flux above a Gmelin larch forest on continuous permafrost in Central Siberia during a growing season, Theor. Appl. Climatol., 93(3–4), 133–147, doi:10.1007/s00704-007-0337-x, 2008.

Oreshkova, N. V., Belokon, M. M. and Jamiyansuren, S.: Genetic diversity, population structure, and differentiation of Siberian larch, Gmelin larch, and Cajander larch on SSR-marker data, Russ. J. Genet., 49(2), 178–186, doi:10.1134/S1022795412120095, 2013.

Paradis, E.: Pegas: An R package for population genetics with an integrated-modular approach, Bioinformatics, 26(3), 419–420, doi:10.1093/bioinformatics/btp696, 2010.

Pluess, A. R.: Pursuing glacier retreat: genetic structure of a rapidly expanding Larix decidua population, Mol. Ecol., 20(3), 473–485, doi:10.1111/j.1365-294X.2010.04972.x, 2011.

Schuelke, M.: An economic method for the fluorescent labeling of PCR fragments A poor man’s approach to genotyping for research and high-throughput diagnostics., Nat. Biotechnol., 18(February), 233–234, 2000.

Shugart, H. H., Leemans, R. and Bonan, G. B.: A Systems Analysis of the Global Boreal Forest, edited by H. H. Shugart, R. Leemans, and G. B. Bonan, Cambridge University Press, Cambridge, UK., 1992.

14

Wagner, S., Gerber, S. and Petit, R. J.: Two highly informative dinucleotide SSR multiplexes for the conifer Larix decidua (European larch)., Mol. Ecol. Resour., 12(4), 717–25, doi:10.1111/j.1755-0998.2012.03139.x, 2012.

Zhang, T., Frauenfeld, O. W., Serreze, M. C., Etringer, A., Oelke, C., McCreight, J., Barry, R. G., Gilichinsky, D., Yang, D., Ye, H. and others: Spatial and temporal variability in active layer thickness over the Russian Arctic drainage basin, J. Geophys. Res., 110(D16), D16101, doi:10.1029/2004JD005642, 2005.